Apply the qwen3_5_quantized_embedding patch: construct the embedding
through the model's quant config so quantized (e.g. fp8/int8)
embeddings are supported, and give it a stable prefix for weight
loading.
Co-Authored-By: Hinata <hinata@idihore.id>
Signed-off-by: MOVZX <movzx@yahoo.com>
